Delineating imprecise regions via shortest-path graphs

An imprecise region, also called a vernacular region, is a region without a precise or administrative boundary. We present a new method to delineate imprecise regions from a set of points that are likely to lie inside the region. We use shortest-path graphs based on the squared Euclidean distance which capture the shape of region boundaries well. Shortest-path graphs naturally adapt to point sets of varying density, and they are always connected. As opposed to neighborhood graphs, they use a non-local criterion to determine which points to connect. Furthermore, shortest- path graphs can easily be extended to take geographic context into account by modeling context as "soft" obstacles. We present efficient algorithms to compute shortest-path graphs with or without geographic context. We experimentally evaluate the quality of the imprecise regions computed with our method. To fairly compare our results to those obtained by the common KDE approach, we also show how to integrate context into KDE by again using soft obstacles.